Anthropic says unreleased Claude model found and exploited flaws in Chrome, Safari, Android, Windows
For the CISO of a regional US bank
Anthropic said in a research post published Oct. 27 that an unreleased internal model found critical vulnerabilities in browsers and operating systems and wrote working exploit chains for five attacks.1
Anthropic's Frontier Red Team said it used the model over the past few months to test vulnerability research on widely used software, and that it reported each flaw to the affected vendor under its coordinated disclosure policy.1 That policy gives developers 90 days to patch and a further 45 days before Anthropic discusses details beyond the public advisory.1
In Chrome, Anthropic said Claude chained flaws in the V8 JavaScript engine, CacheStorage and FedCM to run arbitrary code outside the sandbox; its proof of concept launched Calculator on macOS within seconds of a page loading.1
Anthropic said it reported the Chrome findings to Google, and 62 CVEs have been assigned and patched as of Chrome 156, all rated critical or high, including 13 sandbox escapes and 21 vulnerabilities that could have compromised the renderer.1
In Safari, Anthropic said the chain bypassed the same-origin policy, so a site could read history, cookies and other data and send authenticated requests; if a user is logged into a bank account, an attacker could view balances or attempt transfers. Apple acknowledged the report the same day and a fix is available.1
Anthropic said the Android lock screen bypass, tested on a Google Pixel 9, has been addressed by Google, and Microsoft resolved a Windows HTTP/3 (QUIC) denial-of-service flaw assigned CVE-2026-33096; Anthropic estimates there are likely several million such servers on the internet.1
Patching employee devices does not cover customers' browsers, and Anthropic's Safari example shows a signed-in banking session can be a target. Executives should review how they handle disclosure-to-patch timing, browser update rollout and fraud controls for sessions on unpatched customer devices.
Our take. Vendors get 90 days to patch. The Chrome chain ran code seconds after a page loaded, and one model produced 62 Chrome CVEs. Plan as if attackers will have this too, and cut browser patch windows to days.
Action. Endpoint security: Cut forced Chrome and Safari update deadlines on bank devices to days, and confirm every device runs Chrome 156 or later.
What our records say
Anthropic sells banks Claude seats and API access directly and through cloud marketplaces, and in 2026 added embedded engineering with Goldman Sachs and FIS, restricted cyber-defence access to its Mythos model for a few large banks, and a services firm co-founded with Goldman Sachs.1 Among BankingNewsAI's 100 banks, Goldman Sachs and JPMorganChase offer Claude to staff, BankUnited and Flagstar run Claude in internal AI platforms, and several large banks test the restricted Mythos model.1
From our records
- Claude (Anthropic)vendor · banks using it (21) Goldman, JPMorgan, BankUnited · alternatives Microsoft Copilot, ChatGPT Enterprise, Gemini Enterprise